The Agent Skills Directory

[SAFE]: The skill provides legitimate technical documentation and code examples for Godot 4.7 game networking. It correctly emphasizes the "Server Authoritative" model as a security best practice to prevent cheating and unauthorized state changes.
[INDIRECT_PROMPT_INJECTION]: The skill documents mechanisms for processing data from untrusted network peers, representing a standard game development attack surface.
Ingestion points: The auth_callback example in references/replication-and-rpc.md parses UTF-8 strings into JSON from PackedByteArray received over the network.
Boundary markers: The skill does not provide specific prompt boundary markers for the AI, but it relies on standard programmatic validation patterns.
Capability inventory: The skill uses ENetMultiplayerPeer for network communication, MultiplayerSpawner for node instantiation, and @rpc for remote function calls.
Sanitization: The documentation explicitly warns: "Treat all client input as untrusted; keep the server authoritative" and "Never let clients set authoritative state... validate on the server."

godot-multiplayer